
Profilna
Profilna is an AI profile picture generator and it's my own product. You upload twenty photos of yourself, pick twenty styles, and a couple of hours later you get 160 new profile pictures in 4K. I designed it, built it, and still run it today at profilna.rs.
The AI avatar wave hit in November 2022. Weeks later Profilna was live and selling while the trend was still hot.

Why it mattered
In late 2022, Lensa made AI avatars the thing everyone was posting. Demand was massive, the products riding it were thin, and a privacy backlash was brewing over apps that kept people's photos forever. Windows like that close fast, so I made the calls a founder has to make: pick a scope that ships in weeks, get it in front of real paying customers, and earn their trust by treating their photos with respect. I launched in my home market first, where the big tools didn't even have working payments, which meant building the whole commercial side myself.
Experience goals
- Simple enough for anyone. Add your photos, pick your styles, and get an email when your images are ready. There's nothing to install, configure or learn along the way.
- A product with personality. The copy is casual and friendly, and the style catalog goes from a supermodel and a resume photo to a cyborg, a Renaissance king, or Santa.
- Honest about what AI can do. The FAQ says upfront that some images will come out weird, which is exactly why you get eight per style, so at least one of each is perfect.
- Privacy as a promise. Your photos and the AI model trained on them are deleted within days, and the product says so plainly, because people were right to worry about where their faces end up.
- Easy to show off. Every finished gallery has a shareable link, so people could send their results to friends without asking anyone to log in.
What the app delivers
- A landing page that sells the whole thing in three steps. Add photos, choose styles, get 160 images a couple of hours later.
- An order flow with guardrails. Exactly twenty photos, exactly twenty styles picked from a visual gallery, with separate style catalogs depending on who's in the photos.
- A catalog of dozens of hand-tuned styles. Superhero, Pixar character, Van Gogh, rock star, knight, Santa. Each one is a carefully crafted prompt, refined until it worked reliably.
- Card payments through a bank's 3-D Secure gateway. A full checkout integrated directly with a bank, built for a market the usual payment providers didn't cover.
- A generation pipeline that runs on its own. Training an AI model on someone's face takes hours, so the whole system works asynchronously and emails you the moment your images are ready.
- A personal gallery with 4K downloads. Browse all 160 images, download favorites, or grab everything at once as a ZIP.
- Two full language versions. Profilna launched in my home market first and got a complete English counterpart at profilna.rs/en.

Challenges
- The trend had a clock on it. AI avatars were hot right then, so I made every decision with speed in mind and shipped a working product while people were still posting theirs. Some corners were cut deliberately, and I knew which ones.
- Taking payments without Stripe. Stripe doesn't operate in Serbia, so I integrated directly with a bank's payment gateway, the kind with signed requests and 3-D Secure verification, and made it feel like a normal modern checkout.
- Jobs that take hours to finish. The AI trains on your photos and then renders twenty styles, each finishing at its own pace. I built the system around completion callbacks, and it copies every finished image into my own storage so results never disappear.
- Prompt quality was the product. Image models in 2022 loved extra fingers and warped faces. Every style went through rounds of tuning, and generating eight images per style meant even a bad batch still gave you keepers.
- Strangers had to trust me with their faces. Uploading twenty photos of yourself and paying upfront takes real trust, so I earned it with plain answers: what it costs, why it isn't free, what happens to your photos, and when they get deleted.
How I approached it
I built Profilna as a Next.js app with React and Tailwind, with Supabase handling accounts, data and image storage. The images come from Stable Diffusion, fine-tuned on each customer's photos through a hosted training service, and transactional email runs through SendGrid. I designed the brand and the interface myself: a dark, cosmic look with a pink-to-purple gradient that fit the playfulness of the product. The build was a string of pragmatic calls. I switched AI providers mid-project when a better fit appeared, and when it was time to add English I duplicated the interface instead of building a translation layer, because shipping mattered more than elegance.
Outcome
Profilna launched in early 2023 and real people paid for it, uploaded their photos, and got their 160 images. The site is live at profilna.rs today. It's also the project I point to when someone asks what I mean by end-to-end: every call here was mine, what to build, what to charge, how it should look, how it should treat its users, and what to cut so it shipped while it mattered.